NN.Floats.FP32

FP32 umbrella import.

import NN.Floats.FP32 is what we reach for when we want “float32 semantics for proofs”:

• the canonical binary32-like rounding configuration (fexp32, rnd32, FP32),
• the real-level helper operators (round32/ulp32/eps32, plus unicode aliases),
• per-op absolute error bounds (*_abs_error),
• interval-style enclosure corollaries (*_mem_Icc).
• exact subtraction for nearby representable operands via Sterbenz's lemma.

The TorchLean-specific ≈[t] adapter lives in NN.Proofs.RuntimeApprox.FP32; it is intentionally not part of this numerical umbrella.

Most of the implementation lives under NN/Floats/FP32/* to keep the code navigable.